Llama 3.3 models
4 Llama 3.3 models on Hugging Face, from 49.9B to 70.6B parameters. At the precision each one is published in, the smallest needs about 55.7 GB of VRAM (Llama-3_3-Nemotron-Super-49B-v1_5-FP8, cheapest live fit: RTX PRO 6000) and the largest about 158 GB (Llama-3.3-70B-Instruct, cheapest live fit: RTX A5000). The cheapest way to run Llama-3_3-Nemotron-Super-49B-v1_5-FP8 is $1.38/hr.
Llama 3.3 models
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| Llama-3_3-Nemotron-Super-49B-v1 | 49.9B | BF16 | 111 GB | RTX 5060 Ti × 7 | $0.770/hr |
| Llama-3_3-Nemotron-Super-49B-v1_5 | 49.9B | BF16 | 111 GB | RTX 5060 Ti × 7 | $0.770/hr |
| Llama-3_3-Nemotron-Super-49B-v1_5-FP8 | 49.9B | F8_E4M3 | 55.7 GB | RTX PRO 6000 | $1.38/hr |
| Llama-3.3-70B-Instruct | 70.6B | BF16 | 158 GB | RTX A5000 × 7 | $1.23/hr |
VRAM is for the precision the model is published in, with the same overhead and an 8,192-token context assumed on every page; see the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.
More from Meta
- Llama 3.2 (21 models)
- Llama 3.1 (42 models)
- Llama 3 (22 models)
- Llama 2 (11 models)
- Code Llama (3 models)
- Llama (3 models)
- Llama Guard (4 models)
